Agentforce Coworker: Hands-On Look at Salesforce's New AI Teammate (Beta Review)

Why This Review Matters Now

On May 21, 2026, Salesforce CEO Marc Benioff posted on X calling Agentforce Coworker “your new AI teammate inside every search bar.” The announcement landed during a period of unusually sharp scrutiny for Salesforce AI — Gizmodo had run a piece the same week questioning whether Agentforce was delivering on its promises or amounting to “AI vaporware.” Against that backdrop, a new Beta product announcement deserved more than a press release recap.

Agentforce Coworker entered Beta on May 21 for all existing Agentforce customers, with General Availability targeted for June 15 as part of the Summer ‘26 Release. Most practitioner-level coverage published so far has been thin: news posts confirming the launch exist, but no one has walked through what it actually does, how it differs from what came before, or what the gotchas are for an admin deciding whether to flip the switch.

This review covers all of that. The target audience is Salesforce admins, sales ops managers, and IT leads who need to make a real decision in the next two to three weeks — before GA ships.

Agentforce Coworker vs. Einstein Copilot vs. Previous Agentforce Agents

The landscape has gotten cluttered. Before going further, it is worth establishing how Agentforce Coworker is different from the two things people will most naturally compare it to.

Capability	Einstein Copilot	Previous Agentforce Agents	Agentforce Coworker
Where it lives	Sidebar panel within Salesforce records	Configured flows and channels (email, chat, voice)	The global search bar on every Salesforce page
Trigger model	User opens the sidebar and types a prompt	Inbound event (case, lead, inquiry) triggers agent	Always present; user types in search bar or invokes in Slack/Teams
Data scope	Current record context + org data via permissions	Agent-specific data sources defined at build time	Cross-object CRM data: opps, cases, contacts, activities, workflows
Action capability	Summarize, draft, recommend — soft actions	Can send emails, update records, escalate — hard actions	Can read across CRM + external data and take defined actions
Cross-platform	Salesforce UI only	Salesforce, limited Slack integration	Salesforce, Slack, Teams, ChatGPT
Custom build required	No — on by default with Einstein	Yes — Agent Builder required	Partial — base Coworker is on by default; custom agents require Agent Builder
Credit consumption	Included in Einstein tier	Agentforce credits consumed per interaction	Agentforce credits consumed per interaction
Current status	GA	GA	Beta (GA: June 15, 2026)

The core distinction is scope and presence. Einstein Copilot is a contextual assistant — useful when you are already inside a record and want help with that record. Previous Agentforce agents are discrete automations that run in response to events, configured ahead of time for a specific channel and use case.

Agentforce Coworker is neither of those. It is an ambient, always-available AI that surfaces from the search bar regardless of what page you are on, and that can reason across your entire CRM data model — not just the record in front of you. That is a meaningfully different capability class.

What Agentforce Coworker Actually Does

The clearest demonstration of what Coworker adds comes from the use case Andrew Russo shared on X on May 21, which Benioff amplified. The setup: a sales rep needs to answer a question that requires pulling data from the CRM (open opportunities, customer history) and cross-referencing it with ERP data (order status, inventory, billing). Previously that meant opening three or four systems, copying data into a spreadsheet, and spending 45-60 minutes assembling an answer. With Agentforce Coworker, the rep types a natural language query into the search bar — something like “What is the status of Acme Corp’s open order and how does it affect their renewal opportunity?” — and gets a synthesized answer in seconds.

That is not a demo trick. It is the actual value proposition, and it works because of how Coworker is architected. It has access to the full Salesforce data graph — objects, relationships, field history — and can call out to connected external systems through Agentforce’s Data Cloud integrations. The search-bar entry point matters more than it might seem: it is low-friction enough that reps will actually use it, which is a problem that previous Agentforce configurations solved poorly.

Beyond single-question queries, Coworker supports chained actions. Ask it to find the contact, draft a follow-up email, and log the activity — it can do all three steps in sequence without the user navigating between records. This is where it starts to look less like a search improvement and more like an autonomous work layer.

The cross-platform reach is also real. Coworker surfaces in Slack and Microsoft Teams through the Salesforce app integrations, and a ChatGPT interface is listed as supported, though the Teams and ChatGPT surfaces are more limited in Beta than the native Salesforce experience.

What It Cannot Do Today (Beta Limitations)

Beta means real constraints. Some things the marketing does not make obvious:

Credit transparency is limited. Every Coworker interaction consumes Agentforce credits, and the Beta does not yet surface a clear per-interaction credit cost in the UI. Admins enabling Coworker org-wide without modeling credit consumption first will face surprises at month-end.

The Agent Builder experience is rough. Building custom agents that run natively in Slack or Teams requires the Agent Builder configuration layer, which in Beta is developer-adjacent — not the no-code experience the launch materials suggest. Expect Flow-builder-level complexity at minimum.

External data connections require Data Cloud. The cross-system query capability (the ERP + CRM example) depends on having your external data sources already connected through Salesforce Data Cloud. If your org is not on Data Cloud, Coworker’s scope is limited to native Salesforce objects.

Teams and ChatGPT surfaces are partial. In Beta, the Microsoft Teams integration supports read-only responses — it cannot take write actions from Teams the way it can from the Salesforce native experience. ChatGPT integration is available in preview mode only.

No org-level usage dashboards yet. There is currently no built-in report for tracking Coworker usage, credit draw, or user adoption. You can infer some of this from Agentforce analytics, but dedicated Coworker reporting is listed as a post-GA feature.

Setup: Who Can Access and How to Enable It

Eligibility: Any org with an active Agentforce license can enable Coworker during Beta. No separate purchase or waitlist is required. This includes orgs that have Agentforce for Sales, Agentforce for Service, or the base Agentforce platform license.

Prerequisites before enabling:

1. Agentforce must already be active in your org. If you have not set up Agentforce at all, Salesforce released “Setup with Agentforce” as a GA feature on May 26, 2026 — a simplified onboarding flow available from Setup > Agentforce.
2. You need System Administrator access to toggle the feature.
3. Credit monitoring should be in place before turning Coworker on for end users.

Enabling it:

The official infrastructure setup guide is at developer.salesforce.com/docs/data/agentforce-coworker/guide/agentforce-coworker-turn-on-infrastructure.html. The high-level steps:

• Navigate to Setup > Agentforce > Agentforce Coworker
• Toggle the feature on at org level
• Configure permission sets to control which user profiles have access
• Define the data sources Coworker is authorized to query (critical — do not leave this at the default “all objects”)
• Set up credit usage alerts in the Agentforce credit monitoring panel before granting end-user access

The Real-World Use Case: ERP + Sales Data

The scenario that anchors the Coworker launch is the one Russo demonstrated: a sales rep trying to answer a complex question that spans CRM records and an external ERP system. Here is how it plays out in practice.

A rep is preparing for a renewal call with a customer. They need to know: What is the current order status from the ERP? Are there any open support cases? What is the ARR on the renewal opportunity? What is the customer’s historical NPS? Previously, that meant opening Salesforce for the case and opportunity data, opening the ERP directly for order status, possibly opening a Qualtrics or Gainsight report for NPS — then mentally assembling all of it before the call.

With Agentforce Coworker and Data Cloud connections in place, the rep types one query from the Salesforce search bar. Coworker assembles the answer from the CRM graph and the connected ERP data, surfaces it in a structured summary, and can follow up with actions: draft a call prep note, log a pre-call activity, or flag a support case that needs resolution before the renewal conversation.

Russo’s observation — that this collapsed a 45-60 minute manual process to a few seconds — is credible for orgs with Data Cloud integrations in place. It is worth being clear that the ERP integration piece requires Data Cloud. For orgs running on native Salesforce only, the time savings are real but less dramatic.

Concerns: Data Access, Credit Consumption, and Org Governance

Three concerns warrant detailed treatment before any admin enables Coworker for their organization.

Data access scope. Coworker’s cross-object awareness is its core strength and its core risk. An AI that can reason across your entire CRM data model — including field history, related records, and attached files — has a broad attack surface if access is misconfigured. The default sharing settings in most orgs were not designed with “AI-readable” scope in mind. Before enabling Coworker, audit which objects contain sensitive fields (compensation data, HR notes, legal records, board-level accounts) and explicitly exclude them from Coworker’s object access list.

Agentforce credit consumption. Coworker interactions draw from the same Agentforce credit pool as your other agent configurations. There is no separate Coworker credit bucket, which means enabling Coworker for a large sales team can blow through credits budgeted for automated SDR workflows or service agents. The Beta does not yet surface a clear per-interaction cost in the UI, so the only reliable approach right now is to pilot with a small group, monitor the credit ledger daily, and extrapolate before scaling.

Governance and audit. Enterprise orgs with data residency requirements, regulated data (healthcare, financial), or strict field-level security models need to validate that Coworker’s query execution respects those controls. Salesforce’s documentation states that Coworker honors existing field-level security and sharing rules, but Beta behavior should be verified against your specific org configuration before a production rollout.

Pricing Reality

Agentforce Coworker is not a separately licensed product — but that does not mean it is free.

Cost Element	Detail
Agentforce base license	Required; typically $150-$300/user/month or credit-based pricing depending on contract
Coworker access	Included in Agentforce license during Beta; no additional seat cost
Credit consumption	Each Coworker interaction draws from Agentforce credit pool; rate varies by query complexity (1-3 credits per interaction typical)
Data Cloud	Required for cross-system queries (ERP, external data); separate license, typically $0.25/credit or enterprise contract
Agent Builder customization	Requires Salesforce developer resources to build custom agents; no additional license but implementation cost is real
Credit overages	Billed at standard Agentforce overage rate; get your baseline from contract before enabling

The practical implication: if you are an Agentforce customer planning for GA adoption, model Coworker’s credit consumption now. For a 100-rep sales team doing heavy cross-object queries, expect credit consumption in the range of 1,000-5,000 additional credits per day. At standard overage rates, uncapped production rollout is a meaningful cost line.

Who Should Enable Agentforce Coworker

Good fit:

• Salesforce orgs already paying for Agentforce with headroom in their credit pool
• Sales and service teams doing significant “swivel-chairing” across CRM and external systems
• Orgs with Data Cloud integrations already in place — the ERP + CRM use case is only fully unlocked here
• Admins who want a head start on configuration before the GA rush

Not a good fit:

• Orgs on Salesforce without an Agentforce license — you would need to buy in first
• Teams near their Agentforce credit ceiling — Coworker will accelerate the overage
• Orgs with regulated data (HIPAA, SOC 2 type environments) where Beta behavior needs more GA-level validation before a production deployment
• IT leaders expecting a no-code custom agent experience in Beta — the Agent Builder complexity will disappoint

Bottom Line

Agentforce Coworker is real. The search-bar placement is not a gimmick — it is a genuinely lower-friction entry point than any prior Salesforce AI surface, and the cross-object reasoning capability is a step change from what Einstein Copilot can do within a single record context. The ERP + sales data use case is the right frame for understanding the value: if your reps spend meaningful time assembling cross-system answers before calls, Coworker addresses that directly.

The Beta caveat is also real. Credit consumption transparency is thin, the data access defaults are too broad for most enterprise orgs, and the Agent Builder experience is not yet the no-code environment the marketing implies. Enabling Coworker org-wide before GA based on the launch announcement alone would be a mistake.

The right move is a middle path: enable it in a sandbox this week, run a 2-week pilot with a small group of your highest-complexity users, model credit consumption, lock down object access scope, and then decide whether to go broader before or after June 15. Orgs that go through that process will be ahead of the curve for GA — and will avoid the credit and governance surprises that will catch others flat-footed.

The Gizmodo “AI vaporware” critique of Salesforce AI products has merit in aggregate — there is a pattern of announcement-ahead-of-delivery across the Einstein and Agentforce portfolio. Agentforce Coworker is one of the cases where the delivery is close enough to the promise to take seriously. It earns a conditional recommend.

Rating: 3.8 / 5 for Beta (strong concept, meaningful gaps); target: 4.3 / 5 at GA if credit transparency and Agent Builder UX ship as promised.

Related Articles

• Salesforce Einstein AI: What It Actually Does (And What It Doesn’t)
• TDX 2026: What Salesforce’s Headless 360 and Agentforce Vibes Mean for Enterprise IT
• Best AI Tools for Salesforce Admins in 2026
• G2’s Best Enterprise AI Agents 2026: Who Made the List and What It Means for Your Stack
• AI Governance Framework for IT Teams: A Practical Guide